July will move more cargo through American ports than any month on record, and almost none of it answers to a customer who wanted the goods in July. The Global Port Tracker from the National Retail Federation and Hackett Associates forecasts 2.47 million TEU for the month, edging past the 2.4 million set in May 2022. Ben Hackett’s explanation names no shopper: “Much of this increase reflects frontloading ahead of expected tariff increases.” What the record measures is a deadline. Autumn’s demand forecast has become a bet on when a lawyer files a notice, and the merchandise planning stack is absorbing a tariff risk it was never built to price.

The four months after the record are what give it away. A consumer who wants 3.3% more goods in July does not want 5.7% fewer in September, yet that is precisely what the same forecast projects, with October and November sliding as well. The curve describes the same cargo, counted early: boxes that would have crossed the dock in autumn were pulled into July, and autumn was emptied to pay for it. Front-loading does not create volume. It relocates it in time.

Last year already ran this experiment. Full-year 2025 imports finished at 25.4 million TEU, down 0.3% from 2024, which is what twelve months of tariff panics, pull-forwards and emergency bookings netted out to. The waves were real and the annual total barely moved, because every box hurried forward in spring was a box missing in autumn. A planning system that read 2025’s monthly swings as changing demand was reading a customs calendar and calling it a consumer.

The cost of running for the gap should worry a finance director more than the tariff does. Asia–US West Coast spot rates have climbed 120% since mid-May, which is the premium importers are paying for the privilege of arriving early. The tax the wave is supposedly dodging, meanwhile, is still being collected: the 10% Section 122 surcharge lapses by operation of law at 12:01 a.m. on 24 July, and customs has gone on charging it on every entry until then, so most of the record month is paying the very duty it is running from. That duty is also under a court order: the Court of International Trade already ruled the surcharge unlawful, and CBP keeps collecting it only because the Federal Circuit stayed that ruling pending appeal. The one it is running toward is still a proposal.

A merchant can model a tariff rate; a merchant cannot model a docket.

For apparel, the unknowable part is not the rate but its architecture. USTR’s June proposal sorts sixty economies into two tiers, 10% and 12.5%, and the apparel sourcing map straddles both: Bangladesh, Cambodia, Indonesia and Pakistan sit in the lower band, China in the higher one. Buried inside is a textile mechanism that would let some volume of apparel enter at a reduced rate depending on how much US cotton and US textile input the exporting country buys. Gibson Dunn calls it “one of the more novel” and “least developed” features of the notice, and notes that its “key operational details” are still “to be developed through the comment process.” A cotton T-shirt’s landed cost is now a function of a foreign government’s farm purchases, under a formula nobody has finished writing.

This is the input merchandise planning has no column for. Systems built to forecast autumn buys ingest receipts, on-hand inventory and sell-through, and infer demand from the relationship between them. SAP previewed AI-native retail tools in January that unify sales, inventory, customer and supplier data to push AI into merchandise planning, as TechTarget reported. Two of those four inputs, inventory and supplier data, are being set this quarter by a customs calendar rather than by anything a shopper did. A model that learns the 2026 seasonal curve from 2026 data will learn that Americans want their autumn coats in July.

The strongest objection is that nobody in the building confuses a receipt with a sale. Merchandise planners have separated inbound cargo from point-of-sale data since long before anyone put a model on top of it; port volume is a supply metric, demand comes off the register, and a competent planner reads July’s boxes as inventory position rather than appetite. For the argument here to fail, that separation has to hold all the way down. It does not, because the contamination does not enter through the receipt. It enters through the markdown.

Inventory that lands in July and has not cleared by September gets discounted, and the discount is where policy turns into demand. Cutting the price lifts sell-through, sell-through is the clean signal the forecast trusts, and the model dutifully records that the category sold well at a lower price. Next season’s plan is then built to a price point that a customs deadline set. The receipt was quarantined; the markdown was not. If tariff dates keep pulling inventory forward and margin keeps paying for the storage, then the demand curve retailers plan against will increasingly be a record of their own legal exposure, laundered through a discount.

The choice this forces is narrower than it looks. Either the planning system acquires a variable it cannot observe, a field for what a trade lawyer might do in August, or the merchant overrides the forecast by hand every season and the model becomes expensive decoration. Retailers have spent a decade buying systems that promise to read the customer. This month, the customer is not the one setting the arrival dates.